Replicate application data between servers, browsers and peers. Every write is an op in a persisted log, so peers resume where they stopped, newcomers bootstrap from a snapshot, and every replica converges without a coordinator. Storage, schema and queries are included, and there are no runtime dependencies.
syncom is measured against Yjs, Automerge, Loro, TinyBase, PouchDB, RxDB, Fireproof, GUN and OrbitDB on the same scenarios, in memory and on disk. Each library runs in a fresh process, is used the way its documentation shows, and is checked to give the right answer before it is timed. The chart shows an empty replica pulling a store of 1,000 records; pick another scenario or storage. syncom has the fastest cold start (about 37 ms) and serves about a million reads a second from its cache. Its writes cost more, because each one is signed, logged, filed, catalogued and chained — on disk about 1,500 a second, where the LevelDB-backed libraries do about 20,000 — and a replica verifies every op it receives, which holds a pull in memory to about 25,000 records a second, where Yjs and Loro do 100,000.
Every scenario, the method, the cold start, the bundle sizes and syncom's own benchmarks →